Machine learning: the new frontier for understanding nanoscale materials


Thursday, 2 November, 2017 -
11:00 to 12:00


LG05, Tyree Energy Technologies Building, UNSW Sydney


School of Materials Science and Engineering

Type of event: 


Audience / Guests: 

Public / All

The Materials Genome Initiative (MGI), launched six years ago provided a roadmap towards integration of big-data analytics with high-throughput computations for accelerating materials prediction and design. While practices are readily adopted by the theoretical physics and chemistry communities, these analysis procedures are less known to modern day materials scientists, and are only now rapidly intensifying (and can be paralleled with the rise of open-source data science platforms such as scikit-learn and the Jupyter project). 

In this talk, Dr. Rama K. Vasudevan will introduce the field of machine learning as it applies to imaging data from scanning and electron microscopy, showing how these methods are both superior and necessary as the volume of data acquired at any single sitting is far greater than an individual researcher can manually comprehend. Specific techniques and their applications will be highlighted, including the use of Gaussian Processes for predicting phase diagrams in composition-temperature space, and the use of clustering algorithms to automatically determine structural field-temperature phase diagrams from local measurements. He will further highlight that the use of Bayesian statistics in parallel with full information acquisition can allow for advancements in instrumentation, with specific test cases involving current-voltage measurements and force-distance measurements in scanning probe microscopy. Such advances both allow reconstruction of I-V curves in a scanning mode enabling much greater spectral, temporal and spatial resolution, as well as provide uncertainties on the inferred resistance vectors, overcoming a major limitation of traditional approaches. He will end the talk with how use of physical constraints on the models enable more accurate reconstructions, with examples of understanding interaction energies in iron-chalgoenide superconductors, as well as segregation of cations in a manganite thin film, from atomically resolved imaging data. These methods facilitate the process of turning data from microscopes into knowledge in the form of predictive microscopic models. 

This research was sponsored by the Division of Materials Sciences and Engineering, BES, DOE. This research was conducted at the Center for Nanophase Materials Sciences, which is a US DOE Office of Science User Facility.


Dr. Rama K. Vasudevan is a research and development associate at the Center for Nanophase Materials Sciences, Oak Ridge National Laboratory (ORNL). He received his PhD in Materials Science at UNSW under Prof. V. Nagarajan, before a postdoctoral appointment at ORNL to study manganite thin films with scanning tunneling microscopy. His current research interests include ferroelectrics, oxides, scanning probe microscopy, and machine learning for understanding physics from multidimensional data.